the main html document takes around 1s to load here, and is weighting 152kb, so I had a look at the code, because it seems weird to me this page contains that much data
2021-02-06 03700, 2021
zas
I noticed following block: <script type="application/json">{"aggregatedTags":[],"entity":{"entityType":"release","gid":"354edda3-ef4a-4d94-a0b2-95c4eb7caeaf"},....
2021-02-06 03751, 2021
zas
I extracted it from the page: this json weights 111kb alone (over 152kb)
Right, the list of genres should probably be separated and cached.
2021-02-06 03729, 2021
zas
well, s/probably// ;)
2021-02-06 03714, 2021
prabal
hello, hope everyone is doing good.
2021-02-06 03714, 2021
prabal
2021-02-06 03718, 2021
prabal
Mr_Monkey I saw the proposed project ideas for gsoc. I see you did not mention the solr search one this time. Any particular reason behind this?
2021-02-06 03748, 2021
zas
yvanzo: is this a general issue? I mean are there other json that could be cached but aren't?
2021-02-06 03716, 2021
yvanzo
zas: no I think this is specific to genres implementation
2021-02-06 03750, 2021
yvanzo
zas: genres is editable on every page, whereas other properties are editable on edit pages only
2021-02-06 03719, 2021
zas
the main performance issue with mb website isn't the weight of pages but rather the time needed to generate them, downloads hardly start before 700-800ms after query
2021-02-06 03740, 2021
zas
download times are rather short compared to this waiting time
2021-02-06 03757, 2021
ruaok
pristine___: good morning!
2021-02-06 03723, 2021
pristine___
Good morning
2021-02-06 03733, 2021
zas
yvanzo: it seems I fixed the issue with pink/sir, along cage/burnside. It was apparently a problem between conntrack and docker, resulting from low timeouts set by recent sysctl fixes (ufw was overriding ours, causing other issues). I deployed the change on all servers. The tricky part: only containers were affected and only on certain machines...
2021-02-06 03747, 2021
ruaok
snowy in berlin, I see? how is the weather treating you?
2021-02-06 03758, 2021
zas
Hey ruaok & pristine___
2021-02-06 03712, 2021
pristine___
Yes, I love snow <3. Planning for a walk in the afternoon
2021-02-06 03719, 2021
pristine___
Heya zas
2021-02-06 03727, 2021
ruaok
moin zas!
2021-02-06 03758, 2021
ruaok
pristine___: glad you're happy. I had a friend move to Berlin from Barcelona and come mid-october, she said fuck-it and moved back to BCN. lol
2021-02-06 03720, 2021
ruaok
pristine___: so, now that that monster PR for the recommendation feedback is finally merged, what is next on your list?
2021-02-06 03722, 2021
pristine___
lol
2021-02-06 03744, 2021
ruaok
incorporating that feedback into recommendions?
2021-02-06 03705, 2021
pristine___
haha, a very small PR, so that tracks in BP also show up with feedback emoticons
2021-02-06 03722, 2021
pristine___
> incorporating that feedback into recommendions?
2021-02-06 03709, 2021
pristine___
we can do that, though I feel even if get the structure ready we don;t have much data, I am not sure how much people use the feedback system for recommendations
2021-02-06 03742, 2021
ruaok
I think its a chicken and the egg problem. if the system doesn't work people will be frustrated by it.
2021-02-06 03758, 2021
ruaok
we should implement it, so that it can work. then we can tell people to start doing it.
2021-02-06 03735, 2021
pristine___
I am not sure which system you are talking about? are you talking about using feedbacks in recommendations to filter them?
2021-02-06 03736, 2021
ruaok
and for one, I want to never ever be recommended certain artists ever again. I'll spend the time and downvote every single one of their tracks if need be. :)
2021-02-06 03747, 2021
pristine___
:p
2021-02-06 03708, 2021
ruaok
> are you talking about using feedbacks in recommendations to filter them?
2021-02-06 03709, 2021
ruaok
yes
2021-02-06 03738, 2021
pristine___
okay so one thing
2021-02-06 03716, 2021
pristine___
I don't think people use the tracks you might like page, they use these I think, https://listenbrainz.org/user/Vansika%20Pareek/re…, so we should have feedback emoticons here to collect more data. the tracks you might like page has around 2k recs with repetitions, no one will want to re visit it and that is the only page here we have feedback system
2021-02-06 03710, 2021
ruaok
agreed. the tracks you might like should go away pretty soon.
2021-02-06 03720, 2021
pristine___
also, are you planning to have a recommendation project this summer?,might be a good idea. And maybe we can find someone to work after gsoc too
2021-02-06 03728, 2021
pristine___
> agreed. the tracks you might like should go away pretty soon.
2021-02-06 03739, 2021
ruaok
my goal for week after next (I hope!) is to start running troi-bot on a daily basis and making daily jams for everyone I can.
2021-02-06 03720, 2021
ruaok
so that a stready stream of recommendations starts appearing. and the recommendation tab needs to have the feedback in place and then the feedback needs to be taken into account when generating new recs.
2021-02-06 03736, 2021
ruaok
> also, are you planning to have a recommendation project this summer?
2021-02-06 03756, 2021
ruaok
Possibly. I have a feeling too many people will want to do it and few will have the right ideas.
2021-02-06 03714, 2021
pristine___
hmm, I think next step is to organise the recommendations on the site, rn tbh I am somwhat confused which page gives me what, I would what to make it simpler, delete the tracks you might like and have one page for recs maybe
2021-02-06 03725, 2021
pristine___
> Possibly. I have a feeling too many people will want to do it and few will have the right ideas.
2021-02-06 03734, 2021
ruaok
I think we should work on infrastructure -- once our tools are good enough to build troi plugins, many people will likely play with that.
2021-02-06 03749, 2021
pristine___
yeah, but I would say it will good to have a project this summer.
2021-02-06 03751, 2021
pristine___
hmm
2021-02-06 03707, 2021
pristine___
so what do you want to do next, at least have feedbacks on the playlist
2021-02-06 03734, 2021
ruaok
ok, for UI revamp... yes, lots of stuff needs to improve. but, your strengths are on the spark/recs side. I can have Mr_Monkey and/or _lucifer help on the UI bits.
2021-02-06 03702, 2021
pristine___
hmm
2021-02-06 03703, 2021
ruaok
> at least have feedbacks on the playlist
2021-02-06 03738, 2021
ruaok
this is a small PR, yes? maybe start with that. but in general I think your skills are best used in spark land. if you tell me what things you wish to see on the UI, I can coordinate with Mr_Monkey and _lucifer
2021-02-06 03750, 2021
shivam-kapila
Can listen feedback be also used as a troi filter? Possibly more people might be using it. (But it isn't 5 scale feedback)
2021-02-06 03759, 2021
pristine___
yeah, I don't mean to do it myself
2021-02-06 03706, 2021
ruaok
yes, sure.
2021-02-06 03712, 2021
pristine___
I mean that will have be a good thing to have
2021-02-06 03751, 2021
ruaok
pristine___: ok, cool. can you do me a favor and collect your thoughts on what UI changes we should make? en email, a google doc or even jira tickets if things are specific enough?
2021-02-06 03700, 2021
pristine___
yes
2021-02-06 03703, 2021
pristine___
sounds good
2021-02-06 03707, 2021
ruaok
then I can coordinate/plan and we free you up for spark work.
2021-02-06 03726, 2021
pristine___
yup
2021-02-06 03747, 2021
pristine___
so maybe talk about the spark stuff, what you have in mind on that?
2021-02-06 03751, 2021
ruaok
so, then focusing on spark tasks.
2021-02-06 03754, 2021
ruaok
:)
2021-02-06 03757, 2021
pristine___
:)
2021-02-06 03710, 2021
ruaok
first, using the collected feedback in recs. that's very high on my list.
2021-02-06 03739, 2021
ruaok
then our basic recommendations should be more or less "working" no?
2021-02-06 03740, 2021
niceplace has quit
2021-02-06 03702, 2021
pristine___
yeah, I mean the very basic step is to not recommend stuff that people have disliked
2021-02-06 03706, 2021
pristine___
right
2021-02-06 03745, 2021
ruaok
yeah, that is the easy part. recommending tracks and tracks like it that a user has liked would be the next, but harder, step.
2021-02-06 03755, 2021
pristine___
so we need a pipeline/queue to get the feedback data to spark, or do you want to import dumps
2021-02-06 03758, 2021
pristine___
latter maybe
2021-02-06 03718, 2021
ruaok
I think the recs need to get into the spark dumps.
2021-02-06 03748, 2021
ruaok
if you make a ticket that tells me how that should look, then I will do it. possibly as early as this week since I need to dive into dumps for a number of reasons.
2021-02-06 03750, 2021
pristine___
the recs? the once which have a feedback associated to them?
my gut instint is to remove them from the candidate sets.
2021-02-06 03746, 2021
pristine___
use them to train the model, or remove the songs
2021-02-06 03752, 2021
pristine___
o damn
2021-02-06 03703, 2021
pristine___
that was my second suggestion
2021-02-06 03706, 2021
pristine___
:p
2021-02-06 03731, 2021
pristine___
I mean at this point we should just remove them from candidate sets
2021-02-06 03733, 2021
ruaok
I think training the model is good too, but that feels that that is a later step.
2021-02-06 03735, 2021
pristine___
will be easier
2021-02-06 03739, 2021
ruaok
yes, agreed.
2021-02-06 03754, 2021
ruaok
and later when we have a more comprehensive use of recommendations, we should use them in the training process.
2021-02-06 03756, 2021
pristine___
training the model on them/ with them needs some thought and expertise
2021-02-06 03700, 2021
pristine___
yeah
2021-02-06 03702, 2021
ruaok
agreed.
2021-02-06 03706, 2021
pristine___
also
2021-02-06 03744, 2021
pristine___
for using something to train a model, there should be substantial amount of data, the data we have rn on feedback is too less to see its affect on the trained model
2021-02-06 03752, 2021
pristine___
so yeah, removing sound good
2021-02-06 03712, 2021
ruaok
agreed. this allows us to collect data and then later we can use it.
2021-02-06 03722, 2021
pristine___
+1
2021-02-06 03709, 2021
ruaok
then, after this, I'd love to see "artists you might like". So tracks -> artists collaborative filtering.
2021-02-06 03732, 2021
pristine___
ahaan, I think i did some work on that afair
2021-02-06 03737, 2021
ruaok
because that will allow us to make a better similar artists filter for troi and for track based recs.
2021-02-06 03738, 2021
pristine___
I did some cleaning ig
2021-02-06 03746, 2021
ruaok
prep work, yes.
2021-02-06 03709, 2021
pristine___
yeah, it will be similar to tracks you might like so doable
2021-02-06 03710, 2021
ruaok
I think those two pieces are missing for us to cross a basic usability threshold.
2021-02-06 03736, 2021
pristine___
yeah, so I think I have enough for some time on my plate, lol,
2021-02-06 03744, 2021
ruaok
while you work on those, I hope to be working on troi to build basic recommended playlists, updated daily or weekly.
2021-02-06 03714, 2021
ruaok
pristine___: indeed you do. and worry not about UI work. we've got help for that. lets move the spark work forward.
2021-02-06 03724, 2021
pristine___
yes, also, I will surely open the ticket for UI, it is also an imp thing to keep people engage and attract them at the first place
2021-02-06 03729, 2021
pristine___
yeah sure
2021-02-06 03752, 2021
pristine___
just want to pur in my suggestions,and then we have Mr_Monkey <3
2021-02-06 03701, 2021
pristine___
yay
2021-02-06 03710, 2021
ruaok
perfect, sounds good.
2021-02-06 03727, 2021
ruaok
I'm excited to pick up working on this. there have been too many distractions in Jan.
2021-02-06 03749, 2021
pristine___
yeah, for me too, gap of three months almost, lost track on stuff
2021-02-06 03735, 2021
Mr_Monkey
Moin! Just readin backlog and wanted To point something out: the reason we published the playlist feature with love/hate rather than emoticons feedback is that one uses MSIDs. I think the project to use MBIDs instead of MSID should come first
